Single-modal and multi-modal false arrhythmia alarm reduction using attention-based convolutional and recurrent...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Intensive care units (ICUs) face challenges with high rates of false alarms from patient monitoring systems.
• Existing methods for alarm management often rely on rule-based systems or traditional machine learning with manually engineered features, limiting their adaptability and performance.
• Accurate alarm classification is crucial for reducing alarm fatigue among clinicians and ensuring timely intervention.

Purpose of the Study:

• To develop and evaluate a deep learning model for effective suppression of false alarms in ICUs.
• To improve the accuracy of alarm detection using single- and multi-modal biosignals.
• To outperform existing algorithms in reducing false alarms while maintaining high sensitivity for true alarms.

Main Methods:

• Utilized a deep learning architecture combining convolutional neural networks (CNNs) for feature extraction, an attention mechanism for focusing on relevant signal regions, and long short-term memory (LSTM) units for temporal analysis.
• Employed a two-step training strategy involving pre-training and fine-tuning the network.
• Trained the model on the PhysioNet Computing in Cardiology Challenge 2015 dataset, incorporating single- and multi-modal biosignals.

Main Results:

• The proposed deep learning model demonstrated superior performance in false alarm reduction compared to existing methods.
• Achieved a sensitivity of 93.88% and a specificity of 92.05% for alarm classification across three different signals.
• Showcased significant results for specific alarm types, such as Ventricular Tachycardia arrhythmia, with a sensitivity of 90.71%, specificity of 88.30%, and AUC of 89.51 using single-lead ECG.

Conclusions:

• The deep learning model effectively suppresses false alarms in ICUs without compromising the detection of true alarms.
• The integration of CNNs, attention mechanisms, and LSTMs allows for automatic feature extraction and temporal information capture, leading to enhanced alarm classification accuracy.
• This approach offers a promising solution for improving the reliability of patient monitoring systems in critical care settings.